EP2159694B1 - Method and device for barrier synchronization, and multicore processor - Google Patents

Method and device for barrier synchronization, and multicore processor Download PDF

Info

Publication number
EP2159694B1
EP2159694B1 EP07790190.8A EP07790190A EP2159694B1 EP 2159694 B1 EP2159694 B1 EP 2159694B1 EP 07790190 A EP07790190 A EP 07790190A EP 2159694 B1 EP2159694 B1 EP 2159694B1
Authority
EP
European Patent Office
Prior art keywords
processor
barrier
synchronization
logic
core
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
EP07790190.8A
Other languages
German (de)
English (en)
French (fr)
Other versions
EP2159694A4 (en
EP2159694A1 (en
Inventor
Hideyuki Unno
Masaki Ukai
Matthew Depetro
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Fujitsu Ltd
Original Assignee
Fujitsu Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Fujitsu Ltd filed Critical Fujitsu Ltd
Publication of EP2159694A1 publication Critical patent/EP2159694A1/en
Publication of EP2159694A4 publication Critical patent/EP2159694A4/en
Application granted granted Critical
Publication of EP2159694B1 publication Critical patent/EP2159694B1/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/52—Program synchronisation; Mutual exclusion, e.g. by means of semaphores
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/52—Program synchronisation; Mutual exclusion, e.g. by means of semaphores
    • G06F9/522—Barrier synchronisation

Definitions

  • the embodiments discussed herein are related to a technique for realizing barrier synchronization of two or more processor cores or processors belonging to the same synchronization group.
  • Computer systems are being required to achieve higher processing speeds and larger processing volume (throughput) .
  • This increases the importance of distributed processing, which can be realized by using plural processors.
  • distributed processing has to be realized at a high efficiency.
  • Barrier synchronization is one technique that can realize distributed processing at a high efficiency, and is widely used for high-performance computer systems nowadays.
  • barrier synchronization devices for realizing barrier synchronization are disclosed, for example, by Patent Documents 1 through 3.
  • each processor In order to manage the progress of processes in units of synchronization groups, each processor has to report the execution status of a process to other processors, and when each processor has entered a situation in which it has to shift to the next process, the processor also has to report this fact to other processors. Accordingly, distributed processing is roughly classified into two portions; a parallel processing portion and a coordinated operation portion.
  • a parallel processing portion executes process that respective processors have to execute parallelly.
  • a coordinated operation portion is executed in order to make processors operate in a coordinated manner. In order to increase efficiency in distributed processing, it is important to minimize the ratio (time period) needed to execute a coordinated operation portion (synchronization process).
  • multi-core processor having plural processor cores (each processor core having various units for decoding and executing instructions, a registers, cache memory, and the like) each having a computing function.
  • synchronization groups are assigned to individual processor cores. This assignment can bring about a situation in which all processor cores belonging to the same synchronization group are in one multi-core processor. It is thought to be important to take this situation into consideration when a time period taken to execute a coordinated operation portion is to be reduced.
  • Improvement in semiconductor technique has contributed to an increase in processing speed and circuit density of processors and to an increase in capacity of memory. As a result, higher computation performance is realized using an area smaller than realized by conventional techniques, and such areas have become larger. However, the speed of accessing main memory is still lower than the processing speed of a processor. Rather, the gap between processing the speed of processors and operation speed of main memory has become more serious. Accordingly, when some information is transferred via main memory, a period of time taken to execute a coordinated operation portion is influenced by the operation speed of the main memory, which is lower. This point as well has to be taken into consideration in reducing a period of time taken to execute a coordinated operation portion.
  • US 2006/0212868 A1 discloses a barrier synchronization between multiprocessors is performed at high speed while reducing overhead of parallel processing without adding any special hardware mechanism.
  • a parallel computer synchronization method is provided to synchronize threads through barrier synchronization for parallel execution of plural threads on plural processor modules.
  • the parallel computer has plural processor modules equipped with plural processor cores.
  • the processor cores are each assigned plural threads to execute multithread processing.
  • the plural threads are set in hierarchical groups, and barrier synchronization is performed on each group separately.
  • JIAN LI ET AL "The Thrifty Barrier: Energy-Aware Synchronization in Shared-Memory Multiprocessors", HIGH PERFORMANCE COMPUTER ARCHITECTURE, 2004. HPCA-10. PROCEEDINGS. 10 TH INTERNATIONAL SYMPOSIUM ON MADRID, SPAIN 14-18 FEB. 2004, PISCATAWAY, NJ, USA,IEEE, 14 February 2004 (2004-02-14), pages 14-23, XP010778825, DOI: 10.1109/HPCA.2004.10018 ISBN: 978-0-7695-2053-7 discloses a hardware-software approach to saving energy in parallel applications that exhibit barrier synchronization imbalance. Threads that arrive early to a thrifty barrier pick among existing low-power processor sleep states based on predicted barrier stall time and other factors .
  • a barrier synchronization method is a method for synchronizing processes executed by processor cores in a processor having a plurality of processor cores, including a step of assigning, to a same synchronization group, processor cores to be synchronized among the plurality of processor cores, and a step of synchronizing processes executed by the processor cores belonging to the same synchronization group.
  • the barrier synchronization method further include, each of the processor cores included in the processor having a memory, and that the step of synchronizing processes executed by the processor cores include a step of storing, in the memory, synchronization information specifying whether or not the processor cores belonging to the same synchronization group have to shift to a next process. It is desirable that it further include, in addition to the above or instead of the above, a step of making a processor core shift to a quiescent state while a processor core that has completed a process is shifting to a next process when one of the plurality of processor cores has completed a process to be executed.
  • a barrier synchronization method for synchronizing processes executed by processors in an information processing apparatus having a plurality of processors, including a step of making a processor core shift to a quiescent state while a processor core that has completed a process is shifting to a next process when one of the plurality of processors has completed a process to be executed, and a step of returning the processor that was made to shift to the quiescent state to a state before the quiescent state when a timing to shift to a next process has come.
  • a barrier synchronization device which is a barrier synchronization device included in a multi-core processor in order to synchronize at least 2 processor cores belonging to a same synchronization group among a plurality of processor cores included in the multi-core processor, including first information storage means for storing state information representing execution states of processes respectively of the plurality of processor cores, second information storage means for storing combination information representing combinations of processor cores belonging to each of synchronization groups, third information storage means for storing synchronization information representing whether or not the processor cores belonging to the same synchronization group have to shift to a next process, and information updating means for updating, on the basis of the state information and the combination information respectively stored in the first and second information storage means, the synchronization information stored in the third information storage means.
  • the barrier synchronization device further includes, in addition to the configuration of the above device, information writing means for writing, through a dedicated wire, the synchronization information stored in the third information storage means to memory included in the processor cores.
  • N which is the total number of combinations of the second and third information storage means and the information updating means, satisfy: N ⁇ 2 ⁇ M ⁇ X where M represents the total number of the processor cores, and X represents the total number of logic processors of a processor core.
  • a multi-core processor includes a barrier synchronization device for synchronizing at least 2 processor cores belonging to a same synchronization group among the plurality of processor cores, and reporting means for reporting, to the barrier synchronization device, state information indicating execution status of respective processes of the plurality of processor cores.
  • a multi-core processor is implemented with a barrier synchronization device for realizing barrier synchronization, and when two or more processor cores in that multi-core processor belong to the same synchronization group, the barrier synchronization device is used for realizing barrier synchronization of these processor cores.
  • N which is the total number of combinations of the second and third information storage means and the information updating means
  • N which is the total number of combinations of the second and third information storage means and the information updating means
  • M represents the total number of the processor cores
  • X represents the total number of logic processors of a processor core
  • FIG. 1 illustrates a configuration of a multi-core processor (CPU LSI) according to the present embodiment.
  • a processor 1 includes four processor cores 11 (being referred to as "core” in FIG. 1 , and including various units for executing or decoding instructions, registers, cache memory, and the like).
  • Each of the processor cores 11 (hereinafter, referred to as core) is connected to a shared cache control unit 12, and is configured to access or transmit and receive data to and from mass-storage shared cache memory (data port) 13, a bus control unit 14, and a barrier unit 16.
  • a barrier synchronization device according to the present embodiment is realized in the form of the barrier unit 16.
  • the barrier unit 16 When all the cores 11 belonging to the same synchronization group are in the corresponding processor 1, the barrier unit 16 performs a process for realizing barrier synchronization of that synchronization group. Thereby, the necessity of transmitting and receiving data to and from devices out of the processor 1 to realize the barrier synchronization is eliminated. Barrier synchronization is realized in the processor 1 and data transmission/reception, which is slower than the processing speed in the processor, is not performed so that the process for that barrier synchronization can be performed at very high speed.
  • the above multi-core processor (referred to as processor, hereinafter) 1 is designed to be used for building a high-performance computer system.
  • processor hereinafter
  • FIG. 2 such a computer system adopts a configuration in which a plurality of computer nodes 21, each of which serves as a computer system, are connected to a connection device 22 for interconnecting nodes.
  • connection device 22 are crossbars, a mesh network, etc.
  • the above processor 1 adopts a configuration in which all the elements illustrated in FIG.1 are mounted on one LSI.
  • a processor adopting a configuration in which all the elements illustrated in FIG. 1 are included in one package may also be used as the processor 1.
  • a processor adopting a configuration in which at least one semiconductor chip having core 11 mounted thereon and another chip having the shared cache memory 13 mounted thereon are set in one package may also be used.
  • the compute nodes (information processing devices that are computation systems) 21 of FIG. 2 are configured, for example, as illustrated in FIG. 3 .
  • each computer node 21 has plural processors 1, and each of the processors 1 is connected to a system controller 31 through a system bus (not illustrated) .
  • a main storage unit 32 shared by the processors 1 and an input/output control device 33 for inputting and outputting data from and to an external storage device (not illustrated) are connected to the system controller 31.
  • An interface 34 for transmitting and receiving data to and from a connection device 22 is a device that is optionally mounted when the computer node 21 is to be connected to the connection device 22. When the interface 34 is for wireless LANs, the connection device 22 may not be used.
  • FIG. 4 illustrates a configuration of the barrier unit 16.
  • the barrier unit 16 includes 2 barrier processing devices 15 and as many configuration registers 43 as there are logic processors (virtual cores that can be recognized by OS).
  • Each of the barrier processing devices 15 includes a BST group 42 having plural BSTs (Barrier STatus registers), and also includes plural barrier synchronization execution units 40 (referred to as barrier blades in the figure, and will be referred to as "barrier blades 40" hereinafter), each of which can execute a synchronization process of one synchronization group.
  • Each of the BSTs 41 constituting the BST group 42 is assigned a different logic processor.
  • the BSTs 41 are registers for storing one-bit data.
  • Each logic processor core 11 to which that logic processor is assigned
  • the value of the BST 41 represents the execution status of a process of the corresponding logic processor.
  • the point at which the value changes from 0 to 1 indicates the fact that a waiting state has started in order to realize synchronization with another logic processor that belongs to the same synchronization group. This point is referred to as a synchronization point hereinafter.
  • the number of logic processors in each of the cores 11 is two.
  • twenty-four barrier blades 40 in total (3x2x4) are prepared in order to respond to three synchronization groups existing for one logic processor.
  • twenty-four BSTs 41 are prepared as well.
  • FIG. 5 illustrates a configuration of the above barrier blade 40.
  • the barrier blade 40 includes a BST mask register 40a that can hold as many bits as the total number of the BSTs 41, an LBSY (Last Barrier SYnchronization register) 40b, and an LBSY update logic (circuit) 40c.
  • BST mask register 40a that can hold as many bits as the total number of the BSTs 41
  • LBSY Layer SYnchronization register
  • circuit circuit
  • the BST mask register 40a stores combination data representing a combination of logic processors belonging to the synchronization group assigned to the barrier blade 40. For example, the bit values, in combination data, corresponding to a logic processor belonging to the synchronization group and corresponding to a logic processor not belonging to the synchronization group are 1 and 0, respectively.
  • the LBSY update logic 40c uses only the value of the BST 41 assigned to the logic processor belonging to the synchronization group in the BST group 42 on the basis of the BST mask register 40a holding combination data such as described above so that the LBSY update logic 40c can update the value of the LBSY 40b (LBSY value). This update is performed by writing to the LBSY 40b NOT of the current LBSY value (when the current LBSY value is 0, 1 is written and when the value is 1, 0 is written) when the values of all the BSTs 41 as targets are identical.
  • Processes (entire processes of a program including plural threads) that are the target of distributed processing are sometimes subject to change. Specifically, another process may be added or a process may be deleted.
  • That process is assigned the barrier blade 40 for synchronization, and the combination data has to be held in the BST mask register 40a.
  • three barrier blades 40 are prepared for each logic processor.
  • FIG. 9 illustrates a method of updating an LBSY value using the LBSY update logic 40c.
  • "Process 1" through “Process 3" in FIG. 9 respectively represent the processes assigned to logic processors belonging to the same synchronization group, and "BST1" through “BST3” respectively represent the BSTs 41 whose values are updated by the logic processors that execute processes 1 through 3. This is also applied to FIG. 10 , which will be explained later.
  • the value of LBSY 40b is updated from 0 to 1 when the values of all of the BSTs 1 through 3 become 1, and thereafter it is updated from 1 to 0 when the values of all the BSTs 1 through 3 become 0.
  • the cores 11 access various registers in the barrier unit 16, specifically the configuration registers 43, the BSTs 41, the BST mask registers 40a, and the like through the shared cache control unit 12. Thus, accesses take a long period of time to some extent.
  • a logic processor that has written the logical value 1 to the BST 41 after the termination of a process has to monitor (polling) the LBSY value in order to confirm that the value has changed to 1. Because of this monitoring, a spin loop that repeatedly accesses the LBSY value can occur. Accordingly, the present embodiment follows the procedures below, which will be explained specifically by referring to Figs. 6 through 8 and FIG. 10 .
  • FIG. 6 illustrates a mechanism for the writing of the copied value.
  • a selector 61 illustrated in FIG. 6 is prepared for each of the configuration registers 43, i.e., for each logic processor.
  • the LBSY value stored in the LBSY 40b of each of the barrier blades 40 is input into the selectors 61, and each of the selectors 61 selects 6 LBSY values from among the 24 LBSY values so as to output them in accordance with the data stored in the corresponding configuration register 43.
  • the wires on the output side (6 wires in total) are connected to the corresponding core 11 as indicated by the thick lines running through the shared cache control unit 12 in FIG. 7 .
  • the LBSY values selected by the selector 61 are directly output (copied) to the corresponding cores 11.
  • Each of the cores 11 can have 2 logic processors, and accordingly there are 12 wires for one thick line.
  • Each selector 61 is disposed in the barrier unit 16. This is because the closer the selector 61 is disposed to the core 11, the longer the entire wire is. Thus, the entire length of the wire that is needed to input LBSY values to the selector 61 is minimized so as to suppress influence to design and the like.
  • FIG. 8 illustrates a configuration of the core 11 related to the above writing mechanism.
  • the 6 LBSY values output from the writing mechanism are stored in a register 83.
  • AN LBSY value change detection unit 82 can detect change in the LBSY values stored in the register 83.
  • An instruction control unit 81 extracts instructions in the first cache (not illustrated) in order to execute them.
  • the register 83 is provided close to the instruction control unit 81 so that high-speed access can be performed.
  • the register 83 in the core 11 has become a target in which the LBSY values are monitored, fewer monitoring requests are issued to the shared cache control unit 12. Thereby, a higher speed can be achieved, and the load on the shared cache control unit 12 is reduced.
  • Reading LBSY values via the shared cache control unit 12 takes several tens of nano seconds. However, reading data in the register 83 can reduce that period to several nano seconds.
  • Memory onto which LBSY values are to be copied may be of a different type than the register 83.
  • None of the logic processors can execute the process assigned to performing distributed processing until the synchronization point is achieved, i.e., during a period after the value of the corresponding BST 41 is rewritten from 0 to 1 and before the LBSY value is updated (from 0 to 1).
  • a logical processor that has reached the synchronization point shifts to the sleep state.
  • the LBSY value change detection unit 82 detects change in the LBSY values, and is prepared to request that the instruction control unit 81 cancel the sleep state. In response to that sleep state cancelling request, the instruction control unit 81 cancels the sleep state in order to recover to the previous state.
  • Shifting to the sleep state is performed by a program (software) executed by the logic processor.
  • different quiescent states may be adopted as a state to which the instruction control unit 81 shifts after reaching the synchronization point as long as that state can prevent the use of resources (buffer, temporary registers, or like) needed for calculation, can reduce power consumption, and can allow high-speed recovery.
  • FIG. 10 illustrates a period during which it is in the sleep state. Periods during which the logic processors to execute processes 1 through 3 are in the sleep state are displayed as shaded areas of BSTs 1 through 3. As illustrated in FIG. 1 , the respective logic processors are in the sleep state after they have reached the synchronization points, marked by the triangles, and before the LBSY values change to 1.
  • Two pairs each consisting of six registers 83 and one LBSY value change detection unit 82 are prepared. Thereby, even a different logic processor of the core 11 shifts to the sleep state if it has reached the synchronization point, and recovers to the previous state when the LBSY value changes.
  • the changes in the LBSY values can also be determined by the change in the LBSY value corresponding to the synchronization group for which a process was terminated.
  • Shifting to the above sleep state is performed in units of logic processors. However, this may be performed in units of the cores 11, or may be performed in units of one processor. In such a case, the processor 1 does not necessarily need to be a multi-core processor.
  • FIG. 11 illustrates a method of performing barrier synchronization when not all logic processors belonging to the same synchronization group are in one multi-core processor.
  • FIG. 11 two barrier blades 40 are included in the different processors 1.
  • a barrier leaf 110 is included in a constituent element other than the processor 1, such as the system controller 31 illustrated in FIG. 3 or the connection device 22 illustrated in FIG. 2 .
  • the barrier leaf 110 and a BST group 111 are included in the barrier processing device ( FIG. 4 ).
  • the barrier leaf 110 and the like are assumed to be in the barrier device in the connection device 22.
  • Each barrier blade 40 determines the BST 41, in the BST group 42, whose value is to be transmitted by the barrier blade 40 from the BST mask register 40a.
  • the value of the determined BST 41 is transmitted to the connection device 22 through the interface 34, and is stored in the corresponding BST in the BST group 111.
  • the barrier leaf 110 updates the value of the LBSY 110b as necessary on the basis of the combination between BSTs specified by the a mask register 110a in the BST group 111.
  • the value of the LBSY 110b is transmitted to each barrier blade 40, and is stored in the LBSY 40b. By copying the value of the LBSY 40b onto the corresponding core 11, barrier synchronization is realized.
  • LBSY values in the barrier unit 16 can be directly copied onto the respective cores 11 in the present embodiment. However, it is not necessary for all the cores 11 to allow direct copying. It is also possible to employ a configuration in which a sleep state starts when the writing of the value 1 to the corresponding BST 41 starts.

Landscapes

  • Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Memory System Of A Hierarchy Structure (AREA)
  • Multi Processors (AREA)
EP07790190.8A 2007-06-20 2007-06-20 Method and device for barrier synchronization, and multicore processor Active EP2159694B1 (en)

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/JP2007/000664 WO2008155806A1 (ja) 2007-06-20 2007-06-20 バリア同期方法、装置、及びマルチコアプロセッサ

Publications (3)

Publication Number Publication Date
EP2159694A1 EP2159694A1 (en) 2010-03-03
EP2159694A4 EP2159694A4 (en) 2012-12-26
EP2159694B1 true EP2159694B1 (en) 2019-03-27

Family

ID=40155971

Family Applications (1)

Application Number Title Priority Date Filing Date
EP07790190.8A Active EP2159694B1 (en) 2007-06-20 2007-06-20 Method and device for barrier synchronization, and multicore processor

Country Status (4)

Country Link
US (1) US7971029B2 (ja)
EP (1) EP2159694B1 (ja)
JP (1) JP5273045B2 (ja)
WO (1) WO2008155806A1 (ja)

Families Citing this family (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5235870B2 (ja) * 2007-04-09 2013-07-10 パナソニック株式会社 マルチプロセッサ制御装置、その制御方法および集積回路
JP5447807B2 (ja) * 2009-08-07 2014-03-19 株式会社日立製作所 バリア同期方法及び計算機
JP5549574B2 (ja) * 2010-12-17 2014-07-16 富士通株式会社 並列計算機システム、同期装置、並列計算機システムの制御方法
JPWO2012127534A1 (ja) * 2011-03-23 2014-07-24 富士通株式会社 バリア同期方法、バリア同期装置及び演算処理装置
DE102011084569B4 (de) * 2011-10-14 2019-02-21 Continental Automotive Gmbh Verfahren zum Betreiben eines informationstechnischen Systems und informationstechnisches System
US9092272B2 (en) 2011-12-08 2015-07-28 International Business Machines Corporation Preparing parallel tasks to use a synchronization register
JP5974703B2 (ja) * 2012-07-20 2016-08-23 富士通株式会社 情報処理装置およびバリア同期方法
FR3021433B1 (fr) * 2014-05-21 2016-06-24 Kalray Systeme de synchronisation inter-processeurs
GB2549239A (en) * 2014-11-13 2017-10-18 Advanced Risc Mach Ltd Context sensitive barriers in data processing
JP2017016250A (ja) * 2015-06-29 2017-01-19 日本電気株式会社 バリア同期装置、バリア同期方法及びプログラム
US10346164B2 (en) 2015-11-05 2019-07-09 International Business Machines Corporation Memory move instruction sequence targeting an accelerator switchboard
US10152322B2 (en) 2015-11-05 2018-12-11 International Business Machines Corporation Memory move instruction sequence including a stream of copy-type and paste-type instructions
US10140052B2 (en) 2015-11-05 2018-11-27 International Business Machines Corporation Memory access in a data processing system utilizing copy and paste instructions
US9996298B2 (en) 2015-11-05 2018-06-12 International Business Machines Corporation Memory move instruction sequence enabling software control
US10042580B2 (en) 2015-11-05 2018-08-07 International Business Machines Corporation Speculatively performing memory move requests with respect to a barrier
US10126952B2 (en) 2015-11-05 2018-11-13 International Business Machines Corporation Memory move instruction sequence targeting a memory-mapped device
US10241945B2 (en) 2015-11-05 2019-03-26 International Business Machines Corporation Memory move supporting speculative acquisition of source and destination data granules including copy-type and paste-type instructions
US10067713B2 (en) 2015-11-05 2018-09-04 International Business Machines Corporation Efficient enforcement of barriers with respect to memory move sequences
US10318355B2 (en) * 2017-01-24 2019-06-11 Oracle International Corporation Distributed graph processing system featuring interactive remote control mechanism including task cancellation
US11353868B2 (en) 2017-04-24 2022-06-07 Intel Corporation Barriers and synchronization for machine learning at autonomous machines
US10509452B2 (en) * 2017-04-26 2019-12-17 Advanced Micro Devices, Inc. Hierarchical power distribution in large scale computing systems
US11461130B2 (en) 2020-05-26 2022-10-04 Oracle International Corporation Methodology for fast and seamless task cancelation and error handling in distributed processing of large graph data
GB2597078B (en) * 2020-07-14 2022-07-13 Graphcore Ltd Communication between host and accelerator over network
CN112052099A (zh) * 2020-09-03 2020-12-08 上海兆芯集成电路有限公司 微处理器及其处理核心的同步方法

Family Cites Families (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH02144657A (ja) * 1988-11-26 1990-06-04 Hitachi Ltd 並列演算処理装置
JP3285629B2 (ja) 1992-12-18 2002-05-27 富士通株式会社 同期処理方法及び同期処理装置
JPH08187303A (ja) 1995-01-10 1996-07-23 Toyoda Gosei Co Ltd トレーニング装置
JP2783192B2 (ja) 1995-06-21 1998-08-06 日本電気株式会社 バリア同期装置
JP3730740B2 (ja) * 1997-02-24 2006-01-05 株式会社日立製作所 並列ジョブ多重スケジューリング方法
US5940856A (en) * 1997-04-14 1999-08-17 International Business Machines Corporation Cache intervention from only one of many cache lines sharing an unmodified value
JP2003338200A (ja) * 2002-05-17 2003-11-28 Mitsubishi Electric Corp 半導体集積回路装置
JP4276028B2 (ja) 2003-08-25 2009-06-10 株式会社日立製作所 マルチプロセッサシステムの同期方法
US7340565B2 (en) * 2004-01-13 2008-03-04 Hewlett-Packard Development Company, L.P. Source request arbitration
JP4259390B2 (ja) * 2004-04-28 2009-04-30 日本電気株式会社 並列演算処理装置
US7277989B2 (en) * 2004-06-22 2007-10-02 Sun Microsystems, Inc. Selectively performing fetches for store operations during speculative execution
JP4448784B2 (ja) * 2005-03-15 2010-04-14 株式会社日立製作所 並列計算機の同期方法及びプログラム
US7627770B2 (en) * 2005-04-14 2009-12-01 Mips Technologies, Inc. Apparatus and method for automatic low power mode invocation in a multi-threaded processor
JP4471947B2 (ja) * 2005-04-28 2010-06-02 Ｎｅｃエレクトロニクス株式会社 データ処理装置及びデータ処理方法
US7356653B2 (en) * 2005-06-03 2008-04-08 International Business Machines Corporation Reader-initiated shared memory synchronization

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
None *

Also Published As

Publication number Publication date
EP2159694A4 (en) 2012-12-26
US20100095090A1 (en) 2010-04-15
JPWO2008155806A1 (ja) 2010-08-26
US7971029B2 (en) 2011-06-28
JP5273045B2 (ja) 2013-08-28
WO2008155806A1 (ja) 2008-12-24
EP2159694A1 (en) 2010-03-03

Similar Documents

Publication Publication Date Title
EP2159694B1 (en) Method and device for barrier synchronization, and multicore processor
US9971635B2 (en) Method and apparatus for a hierarchical synchronization barrier in a multi-node system
US8549196B2 (en) Hardware support for software controlled fast multiplexing of performance counters
US8769034B2 (en) Query performance data on parallel computer system having compute nodes
KR101400286B1 (ko) 다중 프로세서 시스템에서 작업을 이동시키는 방법 및 장치
US7552312B2 (en) Identifying messaging completion in a parallel computer by checking for change in message received and transmitted count at each node
US7788334B2 (en) Multiple node remote messaging
US8788879B2 (en) Non-volatile memory for checkpoint storage
US7802025B2 (en) DMA engine for repeating communication patterns
EP2778946A2 (en) Hybrid programmable many-core device with on-chip interconnect
US20120297216A1 (en) Dynamically selecting active polling or timed waits
US8065681B2 (en) Generic shared memory barrier
CN108664116B (zh) 网络功能虚拟化的自适应省电方法、装置及cpu控制器
CN104094235A (zh) 多线程计算
US20080178177A1 (en) Method and Apparatus for Operating a Massively Parallel Computer System to Utilize Idle Processor Capability at Process Synchronization Points
Fei et al. FlexNFV: Flexible network service chaining with dynamic scaling
CN117176811B (zh) 阻塞监听多客户并控制多硬件的服务端架构、系统及方法
JP2010134698A (ja) 情報処理システム
CN105718349B (zh) 跨管芯接口监听或全局观察消息排序
JP4117621B2 (ja) データ一括転送装置
JP6688240B2 (ja) 分散同期処理システムおよび分散同期処理方法
CN117149471B (zh) 通信方法、装置、嵌入式系统、存储介质以及电子设备
ITRIQ et al. ADAPTIVE DYNAMIC RESOURCE SYNCHRONIZATION DISTRIBUTED MUTUAL EXCLUSION ALGORITHM (ADRS).
CN116680296A (zh) 一种基于单机的大规模图数据处理系统
Guo et al. Research on inter-process communication based on multi-core MCUs

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

17P Request for examination filed

Effective date: 20100113

AK Designated contracting states

Kind code of ref document: A1

Designated state(s): AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HU IE IS IT LI LT LU LV MC MT NL PL PT RO SE SI SK TR

AX Request for extension of the european patent

Extension state: AL BA HR MK RS

DAX Request for extension of the european patent (deleted)
A4 Supplementary search report drawn up and despatched

Effective date: 20121127

RIC1 Information provided on ipc code assigned before grant

Ipc: G06F 9/52 20060101AFI20121121BHEP

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: EXAMINATION IS IN PROGRESS

17Q First examination report despatched

Effective date: 20170524

GRAP Despatch of communication of intention to grant a patent

Free format text: ORIGINAL CODE: EPIDOSNIGR1

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: GRANT OF PATENT IS INTENDED

INTG Intention to grant announced

Effective date: 20181016

GRAS Grant fee paid

Free format text: ORIGINAL CODE: EPIDOSNIGR3

GRAA (expected) grant

Free format text: ORIGINAL CODE: 0009210

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE PATENT HAS BEEN GRANTED

AK Designated contracting states

Kind code of ref document: B1

Designated state(s): AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HU IE IS IT LI LT LU LV MC MT NL PL PT RO SE SI SK TR

REG Reference to a national code

Ref country code: GB

Ref legal event code: FG4D

REG Reference to a national code

Ref country code: CH

Ref legal event code: EP

REG Reference to a national code

Ref country code: AT

Ref legal event code: REF

Ref document number: 1113869

Country of ref document: AT

Kind code of ref document: T

Effective date: 20190415

REG Reference to a national code

Ref country code: IE

Ref legal event code: FG4D

REG Reference to a national code

Ref country code: DE

Ref legal event code: R096

Ref document number: 602007057962

Country of ref document: DE

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: FI

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

Ref country code: SE

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

Ref country code: LT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

REG Reference to a national code

Ref country code: NL

Ref legal event code: MP

Effective date: 20190327

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: NL

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

Ref country code: BG

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190627

Ref country code: GR

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190628

Ref country code: LV

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

REG Reference to a national code

Ref country code: AT

Ref legal event code: MK05

Ref document number: 1113869

Country of ref document: AT

Kind code of ref document: T

Effective date: 20190327

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: EE

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

Ref country code: SK

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

Ref country code: IT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

Ref country code: PT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190727

Ref country code: RO

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

Ref country code: CZ

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

Ref country code: ES

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: PL

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: AT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

Ref country code: IS

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190727

REG Reference to a national code

Ref country code: DE

Ref legal event code: R097

Ref document number: 602007057962

Country of ref document: DE

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: DK

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

Ref country code: MC

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

REG Reference to a national code

Ref country code: CH

Ref legal event code: PL

PLBE No opposition filed within time limit

Free format text: ORIGINAL CODE: 0009261

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: NO OPPOSITION FILED WITHIN TIME LIMIT

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: SI

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

26N No opposition filed

Effective date: 20200103

REG Reference to a national code

Ref country code: BE

Ref legal event code: MM

Effective date: 20190630

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: TR

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: IE

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20190620

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: BE

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20190630

Ref country code: CH

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20190630

Ref country code: LU

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20190620

Ref country code: LI

Free format text: LAPSE BECAUSE OF NON-PAYMENT OF DUE FEES

Effective date: 20190630

REG Reference to a national code

Ref country code: DE

Ref legal event code: R082

Ref document number: 602007057962

Country of ref document: DE

Representative=s name: HL KEMPNER PATENTANWAELTE, SOLICITORS (ENGLAND, DE

Ref country code: DE

Ref legal event code: R082

Ref document number: 602007057962

Country of ref document: DE

Representative=s name: HL KEMPNER PATENTANWALT, RECHTSANWALT, SOLICIT, DE

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: CY

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

PG25 Lapsed in a contracting state [announced via postgrant information from national office to epo]

Ref country code: MT

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT

Effective date: 20190327

Ref country code: HU

Free format text: LAPSE BECAUSE OF FAILURE TO SUBMIT A TRANSLATION OF THE DESCRIPTION OR TO PAY THE FEE WITHIN THE PRESCRIBED TIME-LIMIT; INVALID AB INITIO

Effective date: 20070620

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: GB

Payment date: 20240521

Year of fee payment: 18

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: DE

Payment date: 20240521

Year of fee payment: 18

PGFP Annual fee paid to national office [announced via postgrant information from national office to epo]

Ref country code: FR

Payment date: 20240522

Year of fee payment: 18